Research On Image Fusion Technology Based On Wavelet Transforms

With the development of information technologies, image fusion becomes indispensable in image processing and understanding. It aims at getting more exact and more comprehensive image description of the same scene or object by distilling and integrating the information of multi-source image. Image fusion technology has been widely applied to remote sensing, medicine, computer vision, military, and so on.The major research of this paper focuses on pixel-level image fusion. The emphasis is image registration and image fusion based wavelet transformation. At first, the paper introduces the background, significance and the present research status at home and abroad. Then the paper expounds traditional image fusion principles, image fusion systems and wavelet transformation theories and methods in detail. Finally, the general evaluation of image fusion parameters is introduced.A prerequisite of pixel-level image fusion is that the source image in space must be strictly aligned, the accuracy of image registration seriously affect the image fusion results. In this paper, the image registration of the principles and basic methods have been studied, the mutual information was used to be a scale of similarity to two registered images and the combination of wavelet transformation, an improved image registration algorithm which overcome the disadvantage of slow registration and bad real-time performance in general algorithms is proposed.On the base of consulting many documents, the existing common image fusion methods were analyzed and researched. Aimed at the choice of fusion rule and fusion operator, a novel image fusion algorithm based on wavelet transformation is presented in this paper. For the low-frequency image of wavelet decomposition, by measuring the correlation coefficient and spatial frequency determine the scale coefficient. In addition, for fusion of low frequency sub-image, according to the directive contrast, the wavelet coefficient of the fused image can be obtained. This algorithm is tested by some images, the experiment shows that this method is feasible and effective.